Author: Christophe Theron
Date: 12:18:49 11/27/00
Go up one level in this thread
On November 26, 2000 at 13:08:36, Ed Schröder wrote:
>On November 26, 2000 at 11:00:08, Christophe Theron wrote:
>
>>On November 25, 2000 at 02:47:29, Uri Blass wrote:
>>
>>>On November 24, 2000 at 23:22:04, Christophe Theron wrote:
>>>
>>>>On November 24, 2000 at 17:02:11, Uri Blass wrote:
>>>>
>>>>>You can see in the Rebel board that the problem of infinite time is known.
>>>>>
>>>>>I found the problem when I analyzed a position for a long time.
>>>>>
>>>>>My experience is that the time is dependent on the position and I had examples
>>>>>when the time was less than 3 hours on my PIII450(192 mbyte hash)
>>>>>
>>>>>I had also an example when the time per move was more than 6 hours and in this
>>>>>case I did not try to check if the inifinite is finite.
>>>>>
>>>>>Uri
>>>>
>>>>
>>>>
>>>>Yes, I am sorry for this one. I have found the bug now, and it works OK. I have
>>>>a computer here that is computing on your position, Uri, since 69 hours and 30
>>>>minutes and is still thinking (BTW it wants to play h4 at depth 21, score is
>>>>+0.38, now computing depth 22).
>>>>
>>>>I guess we will provide a free update to solve this issue.
>>>>
>>>>
>>>>
>>>> Christophe
>>>
>>>Thanks for the information.
>>>
>>>The position is from a game in the final of the championship of Israel in
>>>correspondence games.
>>>
>>>GM Abir Har Aven(ICCF rating 2585) is against me.
>>>
>>>Abir (white) started 1.g3 d5 2.f4 h5 3.Nf3
>>>
>>>I already played h4 and it is Abir to move.
>>>I find that tiger plays h4 with the plan h4 Nxh4 Rxh4 but after giving tiger
>>>the position after h4 Nxh4 it wants to play e5 with a different score(more than
>>>+1 after a few minutes).
>>>
>>>I expect Abir to play Nxh4 and I will have to decide if to play e5 or Rxh4.
>>>
>>>Uri
>>
>>
>>
>>Tiger has computed depth depth 22 after 1.g3 d5 2.f4 h5 3.Nf3 and score is now
>>0.44. Main line begins with: h4 Nxh4 e5 Nf3 exf4 gxf4 Ne7 e3 Nf5 d4 Ng3...
>>
>>It has been computing for 105 hours now, so I think the bug is definitely fixed.
>>
>>
>> Christophe
>
>
>105 hours huh... :)
>
>I remember a bug in Rebel (I believe it was Rebel7) which was baptized
>internally as the "24 hour bug". The case was that at random after exactly
>24 hours when the program was launched Rebel stopped to compute further
>without seeing that on the screen. It was already a pain to find this out,
>fixing wasn't so easy too and testing required to be after the screen
>exactly 24 hours after Rebel7 had been started and see if the bug was
>solved now by watching the debug information on the screen. All in all
>it took about a month to fix the "24 hour bug".
>
>But the nightmare wasn't over as a chess program also has something we
>call the permanent brain and also here the 24 hour bug was an issue. Same
>procedure only this time things went faster as the problem history was
>known.
>
>In total 2-3 lines in the program had changed still it caused about
>5-6 weeks to mark the bug as "solved". Just wonderful.
>
>Ed
The "infinite time is finite" problem has been easier to find in Tiger. However
being sure it is fixed has indeed taken several days.
The problem was a 32 bits node counter. This counter was here since the 286
ages, and had never caused any harm. But nowadays, with 900MHz computers and
faster, the 32 bits counter overflows in a few hours, and it caused the program
to stop thinking!!!
In a few years from now, I think I'll have to watch the NPS counter. Could it be
that this 32 bits counter could overflow? That would mean the program would
compute more than 2G nodes per second!!! :)
Christophe
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.